Modulus SDF usage

[cxi1]

Hello,
I have some questions related to the SDF usage in setting up constraints in Modulus.
Take the fpga codes as example:
"

channel walls insulating

def channel_walls_criteria(invar, params):
    sdf = fpga.sdf(invar, params)
    return np.less(sdf["sdf"], -1e-5)

channel_walls = PointwiseBoundaryConstraint(
    nodes=thermal_nodes,
    geometry=channel,
    outvar={"normal_gradient_theta_f": 0},
    batch_size=cfg.batch_size.channel_walls,
    criteria=channel_walls_criteria,
    quasirandom=cfg.custom.quasirandom,
)
thermal_domain.add_constraint(channel_walls, "channel_walls")

# fluid solid interface
def fpga_criteria(invar, params):
    sdf = channel.sdf(invar, params)
    return np.greater(sdf["sdf"], 0)

fluid_solid_interface = PointwiseBoundaryConstraint(
    nodes=thermal_nodes,
    geometry=fpga,
    outvar={
        "diffusion_interface_dirichlet_theta_f_theta_s": 0,
        "diffusion_interface_neumann_theta_f_theta_s": 0,
    },
    batch_size=cfg.batch_size.fluid_solid_interface,
    criteria=fpga_criteria,
    quasirandom=cfg.custom.quasirandom,
)
thermal_domain.add_constraint(fluid_solid_interface, "fluid_solid_interface")

"
What does " np.less(sdf[“sdf”], -1e-5)" in " channel_walls_criteria" and “np.greater(sdf[“sdf”], 0)” in “fpga_criteria”
mean to set up these constraints? Can anyone tell me how to relate the criteria with the constraints here?
Or Any reference on the sdf library and how it’s used?

Thank you,
Ce

[ngeneva]

Hi @cxi1

What does " np.less(sdf[“sdf”], -1e-5)" in " channel_walls_criteria" and “np.greater(sdf[“sdf”], 0)” in “fpga_criteria” mean to set up these constraints? Can anyone tell me how to relate the criteria with the constraints here?

The criteria is defined in the API doc of constaints. Here we are sampling the boundary of a given geometry but want to consider multiple geometry objects when sampling and the criteria parameter allows us to do this.

The channel_walls constriant is imposing the boundary condition that channel walls are insulating. So the np.less(sdf["sdf"], -1e-5) is forcing the sampling to occur on the channels walls but not inside the FGPA which is attached to the wall.

The fluid_solid_interface constraint is imposing the boundary condition of the FPGA interface with the fluid. Thus we want to sample points from the boundary of the fpga geometry object but exclude any points that are on the bottom of the FPGA which intersect the channel (outside the flow). This is what np.greater(sdf["sdf"], 0), makes sure our points are inside the channel walls / in the flow volume.

The geometry image in the user guide can help clear up what the domain looks like.

Or Any reference on the sdf library and how it’s used?

Signed distance function (SDF) is a property of geometry, used in many of our examples for criteria functions but also spatial weighting of the loss (E.g. LDC example). I would recommend just looking through the different examples to see how its typically used.

[johnlaide]

Hi @ngeneva ,

If I understand correctly, the points with negative sdf values indicate the points outside the geometry in modulus. Regarding the channel_wall constraint, it could be considered that we are sampling the boundary of the channel wall, but only utilizing points outside the FPGA geometry with sdf values smaller than “-1e-5”.

[ngeneva]

Yes! You are correct @johnlaide, I got switched up. Thanks for the correction! Updated my response.

[cxi1]

@ngeneva @johnlaide

Thank you both for the great explanations. If I got the message, regarding the “fluid solid interface” constraint,
it could be considered that we are sampling the boundary of the fpga surface, but only utilizing points inside of the channel wall with sdf values greater than 0.